network-change-simulator / docs /DEMO_SCRIPT.md
Graham Paasch
Polish copy and UI labels for demo
a29b2b1

A newer version of the Gradio SDK is available: 6.12.0

Upgrade

Persona & setup (10–15s)

“You’re an infra engineer bringing up a new AI data-center: hundreds of racks, ToR switches, and a spine–leaf fabric. You can’t afford to guess how risky each change is. This MCP server simulates those changes in our MAESTRO lab and gives you a risk score before you touch production.”

Demo flow (60–90s)

  1. Open the Space / name the pieces (5–10s)

    • Show the HF Space.
    • Say: “This Hugging Face Space is just the front-end. Underneath is an MCP server talking to our MAESTRO lab—a GNS3 fabric plus Ansible, Nornir, and pyATS for checks.”
  2. Scenario 1 – Low-risk VLAN staging (20–25s)

    • In the UI:
      • Change type: VLAN
      • Scenario: Stage VLAN on leaf pair (preset leaf_tor_vlan_stage)
      • Click Run Lightning simulation.
    • Narrate: “First scenario: staging a VLAN on a pair of leaf switches—no traffic swing yet. Lightning mode looks at MAESTRO’s health, the size of the change, and our expected blast radius, and gives us a risk score and explanation.”
    • Point to:
      • Low risk score (~14).
      • Pre-checks: fabric healthy, no existing alarms.
      • Post-checks: staged VLAN present, no new issues.
    • One line: “On a healthy fabric, small, localized changes show up as low-risk with blast radius limited to a couple of leafs.”
  3. Scenario 2 – Riskier TOR uplink shutdown (25–30s)

    • Change type: Interface
    • Scenario: Shutdown TOR uplink in redundant pair (preset tor_uplink_shutdown)
    • Click Run Lightning simulation.
    • Narrate: “Now a riskier scenario: shutting down one TOR uplink in a redundant pair. Same MCP call, same lab, but the risk model knows this can break redundancy.”
    • Point to:
      • Medium risk score (~50–60).
      • Pre-checks: MAESTRO health OK.
      • Post-checks: one adjacency lost, new alarm on the TOR; blast radius confined to one rack.
    • One line: “Even though traffic stays up, the risk score jumps, and the explanation tells us exactly why: lost redundancy and new alarms.”
  4. Scenario 3 – BGP fabric neighbor add (20–25s)

    • Change type: BGP neighbor
    • Scenario: Add fabric neighbor on leaf (preset leaf_bgp_fabric_neighbor_add)
    • Click Run Lightning simulation.
    • Narrate: “Finally, a control-plane change: adding a BGP fabric neighbor on a leaf. We treat BGP changes as inherently more sensitive, even when they succeed.”
    • Point to:
      • Medium-ish risk (~37).
      • Checks: MAESTRO health OK, new neighbor established, no lost adjacencies.
    • One line: “Control-plane changes start at a higher risk baseline, but you still get a quick pass/fail signal from the lab in a single MCP call.”
  5. Closing (10–15s)

    • Narrate: “From an agent’s perspective, this is just one MCP tool—simulate_network_change—that returns risk, blast radius, and explanation in seconds. From a human’s perspective, it’s a way to de-risk common changes in a 1.3-GW data-center rollout without ever hitting production.”